PR #343
Where it stands for merge

chat-shell.tsx and upload-files-fn.ts both hardcoded Robert’s WorkOS id "15". Mark’s Zep memory writes + uploaded decks were attributing to Robert’s account. Now wired to useAuth().user.id. Per-user data isolation restored. (See user_id audit)

openui05DispatchFn, dispatchFn (legacy), interviewFn, findTalentInterviewFn, startOutcomeFn, uploadFilesFn. Backend may ignore the field today (Xano discards unknown body keys); Mark wires server-side at his pace.

"$5M-$15M", "B2B SaaS", "Choose File" labels no longer become permanent conversation titles. Live-verified: new conversations now show "New outcome · Series A" prefix instead of raw fragments. 13 unit tests cover the helper. Backend update_conversation_title endpoint still needed from Mark for full fix.

53 dead AE component files (15,981 lines). 4 lib/hook/constants files (613 lines). 28 components across 7 features (3,479 lines). 796 dead CSS lines from anything-engine.css. 15 dead non-component modules (2,419 lines). Reviewer trust: every line in the diff is meaningful. (See dead-code sweep)

8 patterns: TanStack Router search-reducer cast, useRef no-arg TS5+ migration, JSX namespace import, sonner ToastProps style prop, boolean coercions, Zod-inferred type swap, vestigial myEmails prop, Avatar size enum. Plus 2 test type bridges. (See typecheck cleanup)

All in anything-engine.css. All !important declarations on overrides scoped to dead parent classes. Removing the parent classes dropped the warnings.

Frontend now sends authenticated WorkOS user.id on every BFF that touches user state. Xano ignores unknown body keys today, so memory writes still don’t key per-user until Mark adds the field to the endpoint signatures + writes to user_id-keyed Zep collections.

find_acquisition, find_cofounder, find_collaborators, find_deal_flow, find_job, find_speakers, make_purchase, get_advice. Match the backend’s "9 classes stuck at clarify gate" finding from W13. Each needs Mark’s voice-rule prompt to ship cards.

Frontend coerceConversationTitle helper is the mitigation. Real fix re-titles the row after classification + summary land — needs Mark’s endpoint.

ep 8506 prompt missing the deck-first opening move per prompts/find_investors/interview.md. $prompt_find_investors in Xano needs update to ask "Do you have a pitch deck?" first.

FalkorDB graph data gap. No advisor or journalist nodes exist. Mark needs to backfill the dataset or add a gate that skips these classes for v0.

Returns VC firms / accelerators instead of SaaS prospects. Vector search needs entity-type filter in the synth step.

When interview returns ready=true, canvas auto-dispatches. Mark spec: render editable summary card → user reviews/edits → user confirms → THEN dispatch. Proposal at docs/anything-engine/wave-3-loop13-INTERVIEW-GATE-FIX-PROPOSAL-202605111200.md. ~106 net lines. Mirrors awaitingResultCountRef pattern. 3 design questions for Robert before implementation.

viewTitles["anything-engine"]: "Outcomes" AND viewTitles.outcomes: "Outcomes". Visible duplication in the tablist. Decision: rename one, delete the legacy outcomes tab, or keep both with disambiguating subtitles.

No reviews submitted yet. Mark opens PR #343 (https://github.com/M-Pederson/orbiter-frontend/pull/343), reads the changes, approves or requests changes.

Either accept the proposal as-is, modify it, or defer. Defer-to-merge is fine — the current behavior works for the result-count picker case; only the editable-summary case is missing.

Lint + build + tests + Netlify deploy preview all pass at HEAD ef2649c. No CI work remaining.

Risk: low. Verified Xano discards unknown body keys (per W13 XANO-PROBE). All 5 BFFs send user_id additively; existing fields preserved. If an endpoint hard-rejects, fix is a 1-line revert of the user_id addition on that BFF.

Risk: low. Each of the 100 deleted files was verified via grep -F to have zero importers anywhere in src/. Cross-feature deletion (28 files) was verified by typecheck + lint + tests + canvas dogfood. If Mark imports something we deleted, error appears immediately on his side — restore from git history is one command.

Risk: medium-low. 796 dead CSS lines deleted across 4 commits. Each chunk verified by canvas screenshot dogfood. .ae-chat-surface and .ae-chat-shell parent classes had ZERO JSX usage so descendant rules were definitionally unreachable. /chat standalone route uses different selectors (.ae-msg-*, .ae-bg, .ae-grid) that all stayed intact.

Risk: low. Helper is additive — only affects new conversations. Existing conversation titles unchanged. The "New outcome · " prefix is recognizable and doesn’t collide with any existing title pattern.